fix(lint): Resolve Pylint errors in quantizations.py and nnx_wrappers.py

Author: hsuan-lun-chiang

src/maxtext/layers/nnx_wrappers.py

@@ -498,15 +498,17 @@ def maybe_unbox(x):
       paths_str = ""
       for path, _ in unknown_state_flat.items():
         paths_str += f"\n  - {'/'.join(map(str, path))}"
-        
+
         # Dynamically reconstruct the unknown variables
         curr = module
         for p in path[:-1]:
           if not hasattr(curr, p):
             setattr(curr, p, nnx.Module())
           curr = getattr(curr, p)
 
-      warnings.warn(f"Found unknown module paths in incoming state:{paths_str}. Intermediate modules have been reconstructed.")
+      warnings.warn(
+          f"Found unknown module paths in incoming state:{paths_str}. Intermediate modules have been reconstructed."
+      )
 
     nnx.update(module, new_state)
     _refresh_variable_trace_state(module)

src/maxtext/layers/quantizations.py

@@ -38,9 +38,13 @@
 from flax.linen import fp8_ops
 from flax.linen import initializers as flax_initializers
 import flax.linen as nn
+from flax import nnx
+
+from qwix._src import flax_util
 
 from maxtext.common.common_types import DType, Config
 from maxtext.inference.kvcache import KVQuant
+from maxtext.layers import nnx_wrappers
 
 # Params used to define mixed precision quantization configs
 DEFAULT = "__default__"  # default config
@@ -708,18 +712,16 @@ def configure_kv_quant(config):
 
 
 def _apply_linen_module_in_nnx(linen_module_cls, op_id, *args, **kwargs):
+  """Applies a Linen module within an NNX context."""
   try:
-    from qwix._src import flax_util
     parent = flax_util.get_current_module()
-    from flax import nnx
     is_nnx = isinstance(parent, nnx.Module)
-  except Exception:
+  except Exception:  # pylint: disable=broad-exception-caught
     is_nnx = False
 
   if is_nnx:
     attr_name = f"_qwix_fp8_gpu_{op_id}"
     if not hasattr(parent, attr_name):
-      from maxtext.layers import nnx_wrappers
       rngs = getattr(parent, "qwix_rngs", None)
       if rngs is None:
         rngs = nnx.Rngs(0)
@@ -838,11 +840,17 @@ def maybe_quantize_model(model, config):
     if quantization_provider:
       if config.pure_nnx:
         input_shape = (config.micro_batch_size_to_train_on, config.max_target_length)
-        import jax.numpy as jnp
         dummy_tokens = jnp.ones(input_shape, dtype=jnp.int32)
         dummy_positions = jnp.ones(input_shape, dtype=jnp.int32)
         dummy_segment_ids = jnp.ones(input_shape, dtype=jnp.int32)
-        model = qwix.quantize_model(model, quantization_provider, dummy_tokens, dummy_positions, dummy_segment_ids, enable_dropout=False)
+        model = qwix.quantize_model(
+            model,
+            quantization_provider,
+            dummy_tokens,
+            dummy_positions,
+            dummy_segment_ids,
+            enable_dropout=False,
+        )
       else:
         model = qwix.quantize_model(model, quantization_provider)
   return model